resultsPath=file.path(getwd(),"Results")
# Gather parameters from command line
#dir.create(file.path(resultsPath,"cache"), showWarnings=F, recursive=T)
nCores <- parallel::detectCores()#params$nCores
subsetGenes <- params$subsetGenes
subsetCells <- params$subsetCells
resolution <- as.numeric(params$resolution)
root <- getwd()
# Have to setwd via knitr
# knitr::opts_knit$set(root.dir=resultsPath, child.path = resultsPath)
knitr::opts_chunk$set(echo=T, error=T, root.dir = resultsPath
# cache=T, cache.lazy=T
)
# Utilize parallel processing later on
print(paste("**** Utilized Cores **** =", nCores)) ## [1] "**** Utilized Cores **** = 4"
params ## $subsetGenes
## [1] "protein_coding"
##
## $subsetCells
## [1] 500
##
## $resolution
## [1] 0.6
##
## $resultsPath
## [1] "./"
** ./ **
library(Seurat)
library(dplyr)
library(gridExtra)
library(knitr)
library(plotly)
library(ggplot2)
library(reshape2)
library(shiny)
library(ggrepel)
library(DT)
#
# install.packages('devtools')
# devtools::install_github('talgalili/heatmaply')
## Install Bioconductor
# if (!requireNamespace("BiocManager"))
# install.packages("BiocManager")
# BiocManager::install(c("biomaRt"))
library(biomaRt)
# BiocManager::install(c("DESeq2"))
library(DESeq2)
# library(snow); #BiocManager::install("Rmpi") #NOTE: different lib name than install name (snow vs Rmpi)
createDT <- function(DF, caption="", scrollY=500){
data <- DT::datatable(DF, caption=caption,
extensions = list('Buttons','Scroller'),
options = list( dom = 'Bfrtip',
buttons = c('copy', 'csv', 'excel', 'pdf', 'print'),
scrollY = scrollY, scrollX=T, scrollCollapse = T, paging = F,
columnDefs = list(list(className = 'dt-center', targets = "_all"))
)
)
return(data)
}
# Useful Seurat functions
## Seurat::FindGeneTerms() # Enrichr API
## Seurat::MultiModal_CCA() # Integrates data from disparate datasets (CIA version too)
sessionInfo()## R version 3.5.1 (2018-07-02)
## Platform: x86_64-apple-darwin15.6.0 (64-bit)
## Running under: macOS 10.14.2
##
## Matrix products: default
## BLAS: /Library/Frameworks/R.framework/Versions/3.5/Resources/lib/libRblas.0.dylib
## LAPACK: /Library/Frameworks/R.framework/Versions/3.5/Resources/lib/libRlapack.dylib
##
## locale:
## [1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
##
## attached base packages:
## [1] parallel stats4 stats graphics grDevices utils datasets
## [8] methods base
##
## other attached packages:
## [1] DESeq2_1.22.2 SummarizedExperiment_1.12.0
## [3] DelayedArray_0.8.0 BiocParallel_1.16.5
## [5] matrixStats_0.54.0 Biobase_2.42.0
## [7] GenomicRanges_1.34.0 GenomeInfoDb_1.18.1
## [9] IRanges_2.16.0 S4Vectors_0.20.1
## [11] BiocGenerics_0.28.0 biomaRt_2.38.0
## [13] DT_0.5.1 ggrepel_0.8.0
## [15] shiny_1.2.0 reshape2_1.4.3
## [17] plotly_4.8.0 knitr_1.21
## [19] gridExtra_2.3 dplyr_0.7.8
## [21] Seurat_2.3.4 Matrix_1.2-15
## [23] cowplot_0.9.4 ggplot2_3.1.0
##
## loaded via a namespace (and not attached):
## [1] snow_0.4-3 backports_1.1.3 Hmisc_4.1-1
## [4] plyr_1.8.4 igraph_1.2.2 lazyeval_0.2.1
## [7] splines_3.5.1 digest_0.6.18 foreach_1.4.4
## [10] htmltools_0.3.6 lars_1.2 gdata_2.18.0
## [13] magrittr_1.5 checkmate_1.9.0 memoise_1.1.0
## [16] cluster_2.0.7-1 mixtools_1.1.0 ROCR_1.0-7
## [19] annotate_1.60.0 R.utils_2.7.0 prettyunits_1.0.2
## [22] colorspace_1.3-2 blob_1.1.1 xfun_0.4
## [25] crayon_1.3.4 RCurl_1.95-4.11 jsonlite_1.6
## [28] genefilter_1.64.0 bindr_0.1.1 survival_2.43-3
## [31] zoo_1.8-4 iterators_1.0.10 ape_5.2
## [34] glue_1.3.0 gtable_0.2.0 zlibbioc_1.28.0
## [37] XVector_0.22.0 kernlab_0.9-27 prabclus_2.2-6
## [40] DEoptimR_1.0-8 scales_1.0.0 mvtnorm_1.0-8
## [43] DBI_1.0.0 bibtex_0.4.2 Rcpp_1.0.0
## [46] metap_1.0 dtw_1.20-1 viridisLite_0.3.0
## [49] xtable_1.8-3 progress_1.2.0 htmlTable_1.13.1
## [52] reticulate_1.10 foreign_0.8-71 bit_1.1-14
## [55] proxy_0.4-22 mclust_5.4.2 SDMTools_1.1-221
## [58] Formula_1.2-3 tsne_0.1-3 htmlwidgets_1.3
## [61] httr_1.4.0 gplots_3.0.1 RColorBrewer_1.1-2
## [64] fpc_2.1-11.1 acepack_1.4.1 modeltools_0.2-22
## [67] ica_1.0-2 pkgconfig_2.0.2 XML_3.98-1.16
## [70] R.methodsS3_1.7.1 flexmix_2.3-14 nnet_7.3-12
## [73] locfit_1.5-9.1 tidyselect_0.2.5 rlang_0.3.1
## [76] later_0.7.5 AnnotationDbi_1.44.0 munsell_0.5.0
## [79] tools_3.5.1 RSQLite_2.1.1 ggridges_0.5.1
## [82] evaluate_0.12 stringr_1.3.1 yaml_2.2.0
## [85] npsurv_0.4-0 bit64_0.9-7 fitdistrplus_1.0-11
## [88] robustbase_0.93-3 caTools_1.17.1.1 purrr_0.2.5
## [91] RANN_2.6.1 bindrcpp_0.2.2 pbapply_1.3-4
## [94] nlme_3.1-137 mime_0.6 R.oo_1.22.0
## [97] hdf5r_1.0.1 compiler_3.5.1 rstudioapi_0.9.0
## [100] png_0.1-7 lsei_1.2-0 geneplotter_1.60.0
## [103] tibble_2.0.0 stringi_1.2.4 lattice_0.20-38
## [106] trimcluster_0.1-2.1 pillar_1.3.1 Rdpack_0.10-1
## [109] lmtest_0.9-36 data.table_1.11.8 bitops_1.0-6
## [112] irlba_2.3.2 gbRd_0.4-11 httpuv_1.4.5.1
## [115] R6_2.3.0 latticeExtra_0.6-28 promises_1.0.1
## [118] KernSmooth_2.23-15 codetools_0.2-16 MASS_7.3-51.1
## [121] gtools_3.8.1 assertthat_0.2.0 withr_2.1.2
## [124] GenomeInfoDbData_1.2.0 diptest_0.75-7 doSNOW_1.0.16
## [127] hms_0.4.2 grid_3.5.1 rpart_4.1-13
## [130] tidyr_0.8.2 class_7.3-15 rmarkdown_1.11
## [133] segmented_0.5-3.0 Rtsne_0.15 base64enc_0.1-3
print(paste("Seurat ", packageVersion("Seurat")))## [1] "Seurat 2.3.4"
## ! IMPORTANT! Must not setwd to local path when launching on cluster
# setwd("~/Desktop/PD_scRNAseq/")
dir.create(file.path(root,"Data"), showWarnings=F)
load(file.path(root,"Data/seurat_object_add_HTO_ids.Rdata"))
pbmc <- seurat.obj
rm(seurat.obj)pbmc## An object of class seurat in project RAJ_13357
## 24914 genes across 22113 samples.
metadata <- read.table(file.path(root,"Data/meta.data4.tsv"))
createDT( metadata, caption = "Metadata") ## Warning in instance$preRenderHook(instance): It seems your data is too
## big for client-side DataTables. You may consider server-side processing:
## https://rstudio.github.io/DT/server.html
# Make AgeGroups
makeAgeGroups <- function(){
dim(metadata)
getMaxRound <- function(vals=metadata$Age, unit=10)unit*ceiling((max(vals)/unit))
getMinRound <- function(vals=metadata$Age, unit=10)unit*floor((min(vals)/unit))
ageBreaks = c(seq(getMinRound(), getMaxRound(), by = 10), getMaxRound()+10)
AgeGroupsUniq <- c()
for (i in 1:(length(ageBreaks)-1)){
AgeGroupsUniq <- append(AgeGroupsUniq, paste(ageBreaks[i],ageBreaks[i+1], sep="-"))
}
data.table::setDT(metadata,keep.rownames = T,check.names = F)[, AgeGroups := cut(Age,
breaks = ageBreaks,
right = F,
labels = AgeGroupsUniq,
nclude.lowest=T)]
metadata <- data.frame(metadata)
unique(metadata$AgeGroups)
head(metadata)
dim(metadata)
return(metadata)
}
# metadata <- makeAgeGroups()
pbmc <- AddMetaData(object = pbmc, metadata = metadata)
# Get rid of any NAs (cells that don't match up with the metadata)
if(subsetCells==F){
pbmc <- FilterCells(object = pbmc, subset.names = "nGene", low.thresholds = 0)
} else {pbmc <- FilterCells(object = pbmc, subset.names = "nGene", low.thresholds = 0,
# Subset for testing
cells.use = pbmc@cell.names[0:subsetCells]
)
} Include only subsets of genes by type. Biotypes from: https://useast.ensembl.org/info/genome/genebuild/biotypes.html
subsetBiotypes <- function(pbmc, subsetGenes){
if( subsetGenes!=F ){
print(paste("Subsetting genes:",subsetGenes))
# If the gene_biotypes file exists, import csv. Otherwise, get from biomaRt
if(file_test("-f", file.path(root,"Data/gene_biotypes.csv"))){
biotypes <- read.csv(file.path(root,"Data/gene_biotypes.csv"))
}
else {
ensembl <- useMart(biomart="ENSEMBL_MART_ENSEMBL", host="grch37.ensembl.org",
dataset="hsapiens_gene_ensembl")
ensembl <- useDataset(mart = ensembl, dataset = "hsapiens_gene_ensembl")
listFilters(ensembl)
listAttributes(ensembl)
biotypes <- getBM(attributes=c("hgnc_symbol", "gene_biotype"), filters="hgnc_symbol",
values=row.names(pbmc@data), mart=ensembl)
write.csv(biotypes, file.path(root,"Data/gene_biotypes.csv"), quote=F, row.names=F)
}
# Subset data by creating new Seurat object (annoying but necessary)
geneSubset <- biotypes[biotypes$gene_biotype==subsetGenes,"hgnc_symbol"]
print(paste(dim(pbmc@raw.data[geneSubset, ])[1],"/", dim(pbmc@raw.data)[1],
"genes are", subsetGenes))
# Add back into pbmc
subset.matrix <- pbmc@raw.data[geneSubset, ] # Pull the raw expression matrix from the original Seurat object containing only the genes of interest
pbmc_sub <- CreateSeuratObject(subset.matrix) # Create a new Seurat object with just the genes of interest
orig.ident <- row.names(pbmc@meta.data) # Pull the identities from the original Seurat object as a data.frame
pbmc_sub <- AddMetaData(object = pbmc_sub, metadata = pbmc@meta.data) # Add the idents to the meta.data slot
pbmc_sub <- SetAllIdent(object = pbmc_sub, id = "ident") # Assign identities for the new Seurat object
pbmc <- pbmc_sub
rm(list = c("pbmc_sub","geneSubset", "subset.matrix", "orig.ident"))
}
}
subsetBiotypes(pbmc, subsetGenes)## [1] "Subsetting genes: protein_coding"
## [1] "14827 / 24914 genes are protein_coding"
Filter by cells, normalize , filter by gene variability.
pbmc <- FilterCells(object = pbmc, subset.names = c("nGene", "percent.mito"),
low.thresholds = c(200, -Inf), high.thresholds = c(2500, 0.05))
pbmc <- NormalizeData(object = pbmc, normalization.method = "LogNormalize",
scale.factor = 10000)** Important!**: Specify do.par = T, and num.cores = nCores in ‘ScaleData’ to use all available cores.
# Store the top most variable genes in @var.genes
pbmc <- FindVariableGenes(object = pbmc, mean.function = ExpMean, dispersion.function = LogVMR,
x.low.cutoff = 0.0125, x.high.cutoff = 3, y.cutoff = 0.5)# IMPORTANT!: Must set do.par=T and num.cors = n for large datasets being processed on computing clusters
# IMPORTANT!: Use only the var.genes identified by 'FindVariableGenes' as the 'gene.use' arg in 'ScaleData'
## This will greatly reduced the computational load.
# par.Cores <- ifelse(nCores <= 12, 48, nCores)
pbmc <- ScaleData(object = pbmc, genes.use = pbmc@var.genes, vars.to.regress = c("nUMI", "percent.mito"),
do.par = T, num.cores = nCores)## Regressing out: nUMI, percent.mito
##
## Time Elapsed: 9.55034995079041 secs
## Scaling data matrix
pbmc## An object of class seurat in project RAJ_13357
## 24914 genes across 495 samples.
vp <- VlnPlot(object = pbmc, features.plot = c("nGene", "nUMI", "percent.mito"),nCol = 3, do.return = T) %>% + ggplot2::aes(alpha=0.5)
vp# par(mfrow = c(1, 2))
gp1 <- GenePlot(object = pbmc, gene1 = "nUMI", gene2 = "percent.mito", pch.use=20,
do.hover=T, data.hover = "mut")gp1gp2 <- GenePlot(object = pbmc, gene1 = "nUMI", gene2 = "nGene", pch.use=20,
do.hover=T, data.hover = "mut")gp2ProjectPCA scores each gene in the dataset (including genes not included in the PCA) based on their correlation with the calculated components. Though we don’t use this further here, it can be used to identify markers that are strongly correlated with cellular heterogeneity, but may not have passed through variable gene selection. The results of the projected PCA can be explored by setting use.full=T in the functions above
# Run PCA with only the top most variables genes
pbmc <- RunPCA(object = pbmc, pc.genes = pbmc@var.genes, do.print=F) #, pcs.print = 1:5, genes.print = 5
# Store in Seurat object so you don't have to recalculate it for the tSNE/UMAP steps
pbmc <- ProjectPCA(object = pbmc, do.print=F) VizPCA(object = pbmc, pcs.use = 1:2)PCAPlot(object = pbmc, dim.1 = 1, dim.2 = 2, do.hover=T, data.hover="mut")# 'PCHeatmap' is a wrapper for heatmap.2
PCHeatmap(object = pbmc, pc.use = 1:12, do.balanced=T, label.columns=F, use.full=F) Determine statistically significant PCs for further analysis. NOTE: This process can take a long time for big datasets, comment out for expediency. More approximate techniques such as those implemented in PCElbowPlot() can be used to reduce computation time
#pbmc <- JackStraw(object = pbmc, num.replicate = 100, display.progress = FALSE)
PCElbowPlot(object = pbmc)We first construct a KNN graph based on the euclidean distance in PCA space, and refine the edge weights between any two cells based on the shared overlap in their local neighborhoods (Jaccard similarity). To cluster the cells, we apply modularity optimization techniques such as the Louvain algorithm (default) or SLM [SLM, Blondel et al., Journal of Statistical Mechanics], to iteratively group cells together, with the goal of optimizing the standard modularity function.
On Resolution
The FindClusters function implements the procedure, and contains a resolution parameter that sets the ‘granularity’ of the downstream clustering, with increased values leading to a greater number of clusters. We find that setting this parameter between 0.6-1.2 typically returns good results for single cell datasets of around 3K cells. Optimal resolution often increases for larger datasets. The clusters are saved in the object@ident slot.
# TRY DIFFERENT RESOLUTIONS
pbmc <- StashIdent(object = pbmc, save.name = "pre_clustering")
# pbmc <- SetAllIdent(object = pbmc, id = "pre_clustering")
pbmc <- FindClusters(object = pbmc, reduction.type = "pca", dims.use = 1:10,
resolution = resolution, print.output = F, save.SNN = T,
n.start = 10, nn.eps = 0.5)
PrintFindClustersParams(object = pbmc) ## Parameters used in latest FindClusters calculation run on: 2019-01-11 00:34:18
## =============================================================================
## Resolution: 0.6
## -----------------------------------------------------------------------------
## Modularity Function Algorithm n.start n.iter
## 1 1 10 10
## -----------------------------------------------------------------------------
## Reduction used k.param prune.SNN
## pca 30 0.0667
## -----------------------------------------------------------------------------
## Dims used in calculation
## =============================================================================
## 1 2 3 4 5 6 7 8 9 10
pbmc <- StashIdent(object = pbmc, save.name = "post_clustering") pbmc <- RunUMAP(object = pbmc, dims.use = 1:10, num_threads=0)
# Plot results
DimPlot(object = pbmc, reduction.use = 'umap')As input to the tSNE, we suggest using the same PCs as input to the clustering analysis, although computing the tSNE based on scaled gene expression is also supported using the genes.use argument.
** Important!**: Specify num_threads=0 in ‘RunTSNE’ to use all available cores.
“FItSNE”, a new fast implementation of t-SNE, is also available through RunTSNE. However FItSNE must first be setup on your computer.
labSize <- 6
pbmc <- RunTSNE(object=pbmc, reduction.use = "pca", dims.use = 1:10, do.fast = TRUE,
tsne.method = "Rtsne", num_threads=0) # FItSNE
# note that you can set do.label=T to help label individual clusters
TSNEPlot(object = pbmc, do.label=T, label.size = labSize, do.return=T) %>% ggplotly() %>% toWebGL()tSNE_metadata_plot <- function(var){
print(paste("t-SNE Metadata plot for ", var))
# Metadata plot
p1 <- TSNEPlot(pbmc, do.return = T, do.label = T, group.by = var, pt.size=1,
plot.title=paste("Color by ",var), vector.friendly=T) %>% ggplotly() %>%
layout(legend = list(orientation = 'h', xanchor = "center", x = 0.5, y = .999)) %>% toWebGL()
# t-SNE clusters plot
p2 <- TSNEPlot(pbmc, do.return = T, do.label = T, pt.size=1,
plot.title=paste("Color by Clusters"), vector.friendly=T) %>% ggplotly() %>%
layout(legend = list(orientation = 'h', xanchor = "center", x = 0.5, y = .999)) %>% toWebGL()
#print(plot_grid(ggplotly(p1), ggplotly(p2)))
fluidPage(
fluidRow(
column(6, p1), column(6, p2)
)
)
}
# metaVars <- c(dx","mut","Gender","Age")
#
# for (var in metaVars){
# print(paste("t-SNE Metadata plot for ",var))
# # Metadata plot
# p1 <- TSNEPlot(pbmc, do.return = T, pt.size = 0.5, group.by = var, do.label = T,
# dark.theme=F, plot.title=paste("Color by ",var))
# # t-SNE clusters plot
# p2 <- TSNEPlot(pbmc, do.label = T, do.return = T, pt.size = 0.5, plot.title=paste("Color by t-SNE clusters"))
# print(plot_grid(p1, p2))
# } tSNE_metadata_plot("dx") ## [1] "t-SNE Metadata plot for dx"
## Warning: 'gl' objects don't have these attributes: 'visible', 'showlegend', 'xaxis', 'yaxis', 'hoverinfo', 'type'
## Valid attributes include:
## 'key', 'set', 'frame', 'transforms', '_isNestedKey', '_isSimpleKey', '_isGraticule', '_bbox'
## Warning: 'gl' objects don't have these attributes: 'visible', 'showlegend', 'xaxis', 'yaxis', 'hoverinfo', 'type'
## Valid attributes include:
## 'key', 'set', 'frame', 'transforms', '_isNestedKey', '_isSimpleKey', '_isGraticule', '_bbox'
tSNE_metadata_plot("mut") ## [1] "t-SNE Metadata plot for mut"
## Warning: 'gl' objects don't have these attributes: 'visible', 'showlegend', 'xaxis', 'yaxis', 'hoverinfo', 'type'
## Valid attributes include:
## 'key', 'set', 'frame', 'transforms', '_isNestedKey', '_isSimpleKey', '_isGraticule', '_bbox'
## Warning: 'gl' objects don't have these attributes: 'visible', 'showlegend', 'xaxis', 'yaxis', 'hoverinfo', 'type'
## Valid attributes include:
## 'key', 'set', 'frame', 'transforms', '_isNestedKey', '_isSimpleKey', '_isGraticule', '_bbox'
tSNE_metadata_plot("Gender") ## [1] "t-SNE Metadata plot for Gender"
## Warning: 'gl' objects don't have these attributes: 'visible', 'showlegend', 'xaxis', 'yaxis', 'hoverinfo', 'type'
## Valid attributes include:
## 'key', 'set', 'frame', 'transforms', '_isNestedKey', '_isSimpleKey', '_isGraticule', '_bbox'
## Warning: 'gl' objects don't have these attributes: 'visible', 'showlegend', 'xaxis', 'yaxis', 'hoverinfo', 'type'
## Valid attributes include:
## 'key', 'set', 'frame', 'transforms', '_isNestedKey', '_isSimpleKey', '_isGraticule', '_bbox'
tSNE_metadata_plot("Age") ## [1] "t-SNE Metadata plot for Age"
## Warning: 'gl' objects don't have these attributes: 'visible', 'showlegend', 'xaxis', 'yaxis', 'hoverinfo', 'type'
## Valid attributes include:
## 'key', 'set', 'frame', 'transforms', '_isNestedKey', '_isSimpleKey', '_isGraticule', '_bbox'
## Warning: 'gl' objects don't have these attributes: 'visible', 'showlegend', 'xaxis', 'yaxis', 'hoverinfo', 'type'
## Valid attributes include:
## 'key', 'set', 'frame', 'transforms', '_isNestedKey', '_isSimpleKey', '_isGraticule', '_bbox'
Seurat has several tests for differential expression which can be set with the test.use parameter (see the DE vignette for details). For example, the ROC test returns the ‘classification power’ for any individual marker (ranging from 0 - random, to 1 - perfect).
Shown here: Biomarkers of each cluster vs. all other clusters.
# Limit to only top variable genes?:
# Set arg 'only.pos=F' to capture negative biomarkers
pbmc.markers <- FindAllMarkers(object = pbmc, min.pct = 0.25, thresh.use = 0.25, only.pos = F, test.use = "wilcox")
pbmc.markers <- pbmc.markers %>% mutate(FC = 2^avg_logFC)
createDT(pbmc.markers, caption = paste("All Biomarkers: All Clusters"))topNum = 5
topBiomarkers <- pbmc.markers %>% group_by(cluster) %>% top_n(topNum, avg_logFC)
createDT(pbmc.markers, caption = paste("All Biomarkers: All Clusters"))getTopBiomarker <- function(pbmc.markers, clusterID, topN=1){
df <-pbmc.markers %>%
subset(p_val_adj<0.05 & cluster==as.character(clusterID)) %>%
arrange(desc(avg_logFC))
top_pct_markers <- df[1:topN,"gene"]
return(top_pct_markers)
}
# clust1_biomarkers <- getTopBiomarker(pbmc.markers, clusterID=1, topN=2)
# clust2_biomarkers <- getTopBiomarker(pbmc.markers, clusterID=2, topN=2)
### Plot biomarkers
plotBiomarkers <- function(pbmc, biomarkers, cluster){
biomarkerPlots <- list()
for (marker in biomarkers){
p <- VlnPlot(object = pbmc, features.plot = c(marker), y.log=T, return.plotlist=T)
biomarkerPlots[[marker]] <- p + ggplot2::aes(alpha=0.5) + xlab( "Cluster") + ylab( "Expression")
}
combinedPlot <- do.call(grid.arrange, c(biomarkerPlots, list(ncol=2, top=paste("Top DEG Biomarkers for Cluster",cluster))) )
# biomarkerPlots <- lapply(biomarkers, function(marker) {
# VlnPlot(object = pbmc, features.plot = c(marker), y.log=T, return.plotlist=T) %>% + ggplot2::ggtitle(marker) %>% ggplotly()
# })
# return(subplot(biomarkerPlots) )
}
top1 <- pbmc.markers %>% group_by(cluster) %>% top_n(1, avg_logFC)
nCols <- floor( sqrt(length(unique(top1$cluster))) )
figHeight <- nCols *7
# Plot top 2 biomarker genes for each
for (clust in unique(pbmc.markers$cluster)){
cat('\n')
cat("### Cluster ",clust,"\n")
biomarkers <- getTopBiomarker(pbmc.markers, clusterID=clust, topN=2)
plotBiomarkers(pbmc, biomarkers, clust)
cat('\n')
} ##Construct the plot object
volcanoPlot <- function(DEG_df, caption="", topFC_labeled=5){
DEG_df$sig<- ifelse( DEG_df$p_val_adj<0.05 & DEG_df$avg_logFC<1.5, "p_val_adj<0.05",
ifelse( DEG_df$p_val_adj<0.05 & DEG_df$avg_logFC>1.5, "p_val_adj<0.05 & avg_logFC>1.5",
"p_val_adj>0.05"
))
DEG_df <- arrange(DEG_df, desc(sig))
vol <- ggplot(data=DEG_df, aes(x=avg_logFC, y= -log10(p_val_adj))) +
geom_point(alpha=0.5, size=3, aes(col=sig)) +
scale_color_manual(values=list("p_val_adj<0.05"="turquoise3",
"p_val_adj<0.05 & avg_logFC>1.5"="purple",
"p_val_adj>0.05" = "darkgray")) +
theme(legend.position = "none") +
xlab(expression(paste("Average ",log^{2},"(fold change)"))) +
ylab(expression(paste(-log^{10},"(p-value)"))) + xlim(-2,2) +
## ggrepl labels
geom_text_repel(data= arrange(DEG_df, p_val_adj, desc(avg_logFC))[1:topFC_labeled,],
# filter(DEG_df, avg_logFC>=1.5)[1:10,],
aes(label=gene), color="black", alpha=.5,
segment.color="black", segment.alpha=.5
) +
# Lines
geom_vline(xintercept= -1.5,lty=4, lwd=.3, alpha=.5) +
geom_vline(xintercept= 1.5,lty=4, lwd=.3, alpha=.5) +
geom_hline(yintercept= -log10(0.05),lty=4, lwd=.3, alpha=.5) +
ggtitle(caption)
print(vol)
}
for (clust in unique(pbmc.markers$cluster)){
cat('\n')
cat("### Cluster ",clust,": Volcano")
cap <- paste("Cluster",clust,"DEG Table")
DEG_df <- subset(pbmc.markers, cluster==as.character(clust)) %>% arrange(desc(avg_logFC))
volcanoPlot(DEG_df, caption = cap)
createDT(DEG_df, caption = cap)
cat('\n')
}##
## ### Cluster 0 : Volcano
##
##
## ### Cluster 1 : Volcano
## Warning: Removed 1 rows containing missing values (geom_point).
## Warning: Removed 1 rows containing missing values (geom_text_repel).
##
##
## ### Cluster 2 : Volcano
fp <- FeaturePlot(object = pbmc, features.plot = top1$gene, cols.use = c("grey", "purple"),
reduction.use = "tsne", nCol = nCols, do.return = T)top5 <- pbmc.markers %>% group_by(cluster) %>% top_n(5, avg_logFC)
# setting slim.col.label to TRUE will print just the cluster IDS instead of
# every cell name
DoHeatmap(object = pbmc, genes.use = top5$gene, slim.col.label=T, remove.key=T) %>% ggplotly() %>% toWebGL()RidgePlot(pbmc, features.plot = top1$gene, nCol = nCols, do.sort = F)## Picking joint bandwidth of 0.291
## Picking joint bandwidth of 0.13
## Picking joint bandwidth of 0.0842
Visualize biomarker expression for each cluster, by disease
top2 <- pbmc.markers %>% group_by(cluster) %>% top_n(2, avg_logFC)
sdp <- SplitDotPlotGG(pbmc, genes.plot = top2$gene, cols.use = c("blue","red"),
x.lab.rot = T, plot.legend = T, dot.scale = 8, do.return = T, grouping.var = "dx")The following plots show the absolute expression of each biomarker, as opposed to avg_logFC which is dependent on the expression patterns of other cell types being compared.
markerList <- c("CD14", "FCGR3A")
get_markerDF <- function(pbmc, markerList){
exp <- pbmc@scale.data %>% data.frame()
marker.matrix <- exp[row.names(exp) %in% markerList, ]
marker.matrix$Gene <- row.names(marker.matrix)
markerMelt <- reshape2:::melt.data.frame(marker.matrix, id.vars = "Gene", variable.name = "Cell",value.name = "Expression")
metaSelect <- pbmc@meta.data[,c("barcode", "dx", "mut","post_clustering",
"percent.mito","nGene", "nUMI")]
markerDF <- merge(markerMelt,metaSelect, by.x="Cell", by.y="barcode")
return(markerDF)
}
markerDF <- get_markerDF(pbmc, markerList)
createDT(markerDF, caption = "Known Marker Expression")# Explore expression differences between groups
marker_vs_metadata <- function(markerDF, meta_var){
# Create title from ANOVA summary
ANOVAtitle <- function(markerDF, marker){
nTests <- length(unique(markerDF$Gene))
res <- anova(lm(data = subset(markerDF, Gene==marker),
formula = Expression ~ eval(parse(text=meta_var))))
title <-paste(paste("ANOVA (",marker, " vs. ",meta_var, ")", sep=""),
": p=",round(res$`Pr(>F)`,3),
", F=",round(res$`F value`,3),
ifelse(res$`Pr(>F)`<.05/nTests,"(Significant**)",
"(Non-significant)") )
}
title = ""
for (marker in unique(markerDF$Gene) ){
print(marker)
title <- paste(title, "\n", ANOVAtitle(markerDF, marker))
}
ggplot(markerDF, aes(x=eval(parse(text=meta_var)), y=Expression, fill= Gene)) +
geom_boxplot() +
labs(title = title, x=meta_var) +
theme(plot.title = element_text( size=10)) +
scale_fill_manual(values=c("brown", "slategray"))
}marker_vs_metadata(markerDF, "dx")## [1] "CD14"
## [1] "FCGR3A"
marker_vs_metadata(markerDF, "mut") ## [1] "CD14"
## [1] "FCGR3A"
identify_cellTypes_by_biomarkers <- function(pbmc.markers, topN_search=5){
top <- pbmc.markers %>% group_by(cluster) %>% top_n(topN_search, avg_logFC)
clust_cellTypes <- list()
for (clust in top$cluster){
clustSub <- top[top5$cluster==clust, ]
CD16_logFC <- subset(clustSub, gene=="CFD")$avg_logFC
cellType <- ifelse(sum(markerList %in% clustSub$gene), # Both CD14 and CD16? Great, keep going
ifelse(CD16_logFC == abs(CD16_logFC), "CD14++/CD16+", # But does CD16 have pos logfC? If so, then it's "CD14++/CD16+"
"CD14++/CD16--"), # Otherwise, it means it means CD16 logFC is neg, meaning "CD14++/CD16--"
"N/A") # If it's none of these, it's an undefined cell type
clust_cellTypes[clust] <- cellType
}
newMeta <- pbmc@meta.data
newMeta["CellType_DGE"] <- plyr::mapvalues(metaD$post_clustering, names(clust_cellTypes), as.character(clust_cellTypes) )
pbmc <- AddMetaData(pbmc, metadata = newMeta)
return(pbmc)
}
pbmc <- identify_cellTypes_by_biomarkers(pbmc.markers, 5)## Error in plyr::mapvalues(metaD$post_clustering, names(clust_cellTypes), : object 'metaD' not found
# (Doesn't make sense to do bar plot because whole clusters are defined by their biomarkers)
tSNE_metadata_plot("CellType_DGE")## [1] "t-SNE Metadata plot for CellType_DGE"
## Error in FetchData(object = object, vars.all = group.by): Error: CellType_DGE not found
# A simplistic way of categorizing cells into CD14++/CD16+ and CD14++/CD16--,
## is by splitting cells into groups based on whether their expression is
## higher or lower than the average CD16 expression of all cells.
identify_cellTypes_by_avgExpression <- function(pbmc, markerDF){
avgMarkerExp <-markerDF %>% group_by(Gene) %>% dplyr::summarise(meanExp = mean(Expression))
avgMarkerExp <- setNames(avgMarkerExp$meanExp, avgMarkerExp$Gene)
CD16 <- markerDF[markerDF$Gene=="FCGR3A",]
CD16_group <- ifelse(CD16$Expression >= avgMarkerExp["FCGR3A"], "CD14++/CD16+", "CD14++/CD16--")
CD16["CellType_AvgExp"] <- CD16_group
# Make sure row order is same before putting back into meta.data
metaD <- pbmc@meta.data
newMeta <- merge(metaD, CD16[,c("Cell","CellType_AvgExp")], by.x="barcode", by.y="Cell")
row.names(newMeta) <- row.names(metaD)
pbmc <- AddMetaData(pbmc, metadata = newMeta)
return(pbmc)
}
pbmc <- identify_cellTypes_by_avgExpression(pbmc, markerDF)
# Get proportions of cell types in each cluster
cluster_proportions <- pbmc@meta.data %>% group_by(CellType_AvgExp, post_clustering) %>%
tally() %>%
group_by(post_clustering, CellType_AvgExp) %>%
mutate(percentTotal = n/sum(n)*100)
ggplot(cluster_proportions, aes(x=post_clustering, y=percentTotal, fill=CellType_AvgExp)) + geom_col(position = "fill") +
labs(title="Proportions of Cell-types per Cluster: \n CellType_AvgExp",
x="Cluster", y="Cell Type / Total Cells") +
scale_fill_manual(values=c("brown", "slategray"))tSNE_metadata_plot("CellType_AvgExp")## [1] "t-SNE Metadata plot for CellType_AvgExp"
## Warning: 'gl' objects don't have these attributes: 'visible', 'showlegend', 'xaxis', 'yaxis', 'hoverinfo', 'type'
## Valid attributes include:
## 'key', 'set', 'frame', 'transforms', '_isNestedKey', '_isSimpleKey', '_isGraticule', '_bbox'
## Warning: 'gl' objects don't have these attributes: 'visible', 'showlegend', 'xaxis', 'yaxis', 'hoverinfo', 'type'
## Valid attributes include:
## 'key', 'set', 'frame', 'transforms', '_isNestedKey', '_isSimpleKey', '_isGraticule', '_bbox'
markerDF <- markerDF %>% mutate(Cluster = post_clustering)
# Show mean exp for each marker
avgMarker <- markerDF %>% group_by(Gene, Cluster) %>% summarise(meanExp = mean(Expression))
ggplot(data = avgMarker, aes(x=Gene, y=Cluster, fill=meanExp)) %>% + geom_tile() %>% + scale_fill_distiller(palette="viridis") %>% ggplotly() %>% toWebGL()## Warning in pal_name(palette, type): Unknown palette viridis
# Show mean exp for each marker
avgMarker <- markerDF %>% group_by(Gene, dx, Cluster) %>% summarise(meanExp = mean(Expression))
ggplot(data = avgMarker, aes(x=Gene, y=Cluster, fill=meanExp)) %>% + geom_tile() %>% + scale_fill_distiller(palette="viridis") %>% ggplotly() %>% toWebGL()## Warning in pal_name(palette, type): Unknown palette viridis
markerMelt <- reshape2::acast(markerDF, Cell~Gene, value.var="Expression", fun.aggregate = mean, drop = F, fill = 0)
#plot_ly( z = markerMelt, y=row.names(markerMelt), z=colnames(markerMelt), type="heatmap")
# dx_colors <- colorRampPalette(brewer.pal(2, "RdBu"))
# mut_colors <- colorRampPalette(brewer.pal(length(unique(pbmc@meta.data$mut)), "Set3"))
Spectral <- grDevices::colorRampPalette(RColorBrewer::brewer.pal(length(unique(pbmc@meta.data$mut)), "Spectral"))
# Spectral <- heatmaply::Spectral(length(unique(pbmc@meta.data$mut)))
heatmaply::heatmaply(markerMelt, key.title="Expression",#plot_method= "plotly",
k_row = length(unique(pbmc.markers)), dendrogram = "row",
showticklabels = c(T, F), xlab = "Known Markers", ylab = "Cells", column_text_angle = 0,
row_side_colors = pbmc@meta.data[,c("dx","mut", "CellType_DGE")], row_side_palette = Spectral
) %>% colorbar(tickfont = list(size = 12), titlefont = list(size = 14), which = 2) %>%
colorbar(tickfont = list(size = 12), titlefont = list(size = 14), which = 1)## Error in `[.data.frame`(pbmc@meta.data, , c("dx", "mut", "CellType_DGE")): undefined columns selected
ggplot(data = markerDF, aes(x=Cluster, y=Expression, fill=Gene)) %>%
+ geom_boxplot(alpha=0.5) %>% + scale_fill_manual(values=c("purple", "turquoise")) # %>% ggplotly() expressionTSNE <- function(pbmc, marker, colors=c("grey", "red")){
FeaturePlot(object = pbmc, features.plot = marker, cols.use = colors,
reduction.use = "tsne", nCol=2, do.return = T, dark.theme = T)[[1]] %>% ggplotly() %>% toWebGL()
}
tp1 <- expressionTSNE(pbmc, markerList[1])tp2 <- expressionTSNE(pbmc, markerList[2], colors=c("grey", "green"))subplot(tp1, tp2, titleX = T, titleY = T )library(shiny)
wellPanel( fluidPage(fluidRow(column(6,tp1), column(6,tp2) ),
tags$style("html {background:black;}") ),
style = "background:black;" )current.cluster.ids <- unique(pbmc.markers$cluster) #c(0, 1, 2, 3, 4, 5, 6, 7)
top1 <- pbmc.markers %>% group_by(cluster) %>% top_n(1, avg_logFC)
new.cluster.ids <- top1$gene #c("CD4 T cells", "CD14+ Monocytes", "B cells", "CD8 T cells", "FCGR3A+ Monocytes", "NK cells", "Dendritic cells", "Megakaryocytes")
pbmc@ident <- plyr::mapvalues(x = pbmc@ident, from = current.cluster.ids, to = new.cluster.ids)
TSNEPlot(object=pbmc, do.label=T, pt.size=0.5, do.return=T) %>% ggplotly() %>% toWebGL()# Available DGE methods:
## "wilcox", "bimod", "roc", "t", "tobit", "poisson", "negbinom", "MAST", "DESeq2"
runDGE <- function(pbmc, meta_var, group1, group2, test.use="wilcox"){
#print(paste("DGE_allCells",meta_var,sep="_"))
pbmc <- SetAllIdent(pbmc, id = meta_var)
pbmc <- StashIdent(pbmc, save.name = meta_var)
DEGs <- FindMarkers(pbmc, ident.1=group1, ident.2=group2, test.use=test.use)
DEGs$gene <- row.names(DEGs)
return(DEGs)
}DEG_df <-runDGE(pbmc, "dx", group1 = "PD", group2="control")
cap = paste("DEGs (All Cells): PD vs. Controls")
createDT(DEG_df, caption = cap)volcanoPlot(DEG_df, caption = cap)DEG_df <-runDGE(pbmc, "mut", "LRRK2", "PD")
cap <- paste("DEGs (All Cells): LRRK2 vs. PD")
createDT(DEG_df, caption = cap)volcanoPlot(DEG_df, caption = cap)DEG_df <-runDGE(pbmc, "CellType_DGE", "CD14++/CD16+", "CD14++/CD16--")## Error in WhichCells(object = object, ident = ident.1): Identity : CD14++/CD16+ not found.
cap <- paste("DEGs (All Cells): CD14++/CD16+ vs. CD14++/CD16--")
createDT(DEG_df, caption = cap)volcanoPlot(DEG_df, caption = cap)DGE_within_clusters <- function(pbmc, meta_var, group1, group2){
for (clust in unique(pbmc@ident)){
# Subset cells by cluster
pbmc <- SetAllIdent(pbmc, id = "post_clustering")
pbmc_clustSub <- SubsetData(pbmc, ident.use = clust, subset.raw = T)
cap <- paste("Cluster ",clust,": \n ",group1," vs. ", group2, sep="")
cat('\n')
cat("### ",cap)
# DGE
DEG_df <-runDGE(pbmc_clustSub, meta_var, group1 , group2 )
# Show results
volcanoPlot(DEG_df, caption = cap)
createDT(DEG_df, caption = cap)
cat('\n')
}
}DGE_within_clusters(pbmc, "dx", "PD", "control") ## Error in WhichCells(object = object, ident = ident.use, ident.remove = ident.remove, : Identity : S100A12 not found.
DGE_within_clusters(pbmc, "mut", "LRKK2", "PD")## Error in WhichCells(object = object, ident = ident.use, ident.remove = ident.remove, : Identity : S100A12 not found.
DGE_within_clusters(pbmc, "CellType_DGE", "CD14++/CD16+", "CD14++/CD16--") ## Error in WhichCells(object = object, ident = ident.use, ident.remove = ident.remove, : Identity : S100A12 not found.
DGE_within_clusters(pbmc, "CellType_AvgExp", "CD14++/CD16+", "CD14++/CD16--")## Error in WhichCells(object = object, ident = ident.use, ident.remove = ident.remove, : Identity : S100A12 not found.
If you perturb some of our parameter choices above (for example, setting resolution=0.8 or changing the number of PCs), you might see the CD4 T cells subdivide into two groups. You can explore this subdivision to find markers separating the two T cell subsets. However, before reclustering (which will overwrite object@ident), we can stash our renamed identities to be easily recovered later.
new_resolution <- 3.0
orig_resolution <- paste("resolution",params$resolution,sep="_")
pbmc <- StashIdent(object = pbmc, save.name = orig_resolution)
## Warning in BuildSNN(object = object, genes.use = genes.use, reduction.type
## = reduction.type, : Build parameters exactly match those of already
## computed and stored SNN. To force recalculation, set force.recalc to TRUE.
pbmc <- FindClusters(object = pbmc, reduction.type = "pca", dims.use = 1:10,
resolution = new_resolution, print.output = F)## 4 singletons identified. 33 final clusters.
pbmc <- StashIdent(object = pbmc, save.name = "resolution_3.0")
plot1 <- TSNEPlot(object = pbmc, do.return = TRUE, no.legend = TRUE, do.label = TRUE, label.size=labSize)
plot2 <- TSNEPlot(object = pbmc, do.return = TRUE, group.by = "ClusterNames_0.6",
no.legend = TRUE, do.label = TRUE, label.size=labSize)## Error in FetchData(object = object, vars.all = group.by): Error: ClusterNames_0.6 not found
plot_grid(plot1, plot2)## Error in plot_grid(plot1, plot2): object 'plot2' not found
res3.0_markers <- FindAllMarkers(object = pbmc, min.pct = 0.25, thresh.use = 0.25, only.pos = F, test.use = "wilcox")
FeaturePlot(object = pbmc, features.plot = top1$gene, cols.use = c("green", "blue"))# Set back to orig
pbmc <- SetAllIdent(object = pbmc, id = orig_resolution) # Save results for EACH run (in their respective subfolders)
saveRDS(pbmc, file=file.path(params$resultsPath, "cd14-processed.rds") )